How a leading office products retailer used AI and intelligent automation to reduce Finance review volume by 80% and recover $200K in annual value — while rebuilding trust between stores and corporate.
When the process of receiving goods relies on manual data entry and a physical piece of paper, the question isn't if discrepancies will happen — it's how much they'll cost by the time anyone notices.
In retail, supply chain accuracy is the foundation of financial integrity. When the data captured at the receiving dock doesn't match what was ordered, shipped, or invoiced, every downstream decision — payment runs, vendor reconciliation, inventory counts — is at risk.
For a major national office products retailer operating hundreds of store locations, inventory receiving was a distributed, high-stakes operation. Unlike centralized distribution models, this company used a ship-to-store approach — suppliers delivered directly to individual store locations, meaning receiving accuracy depended on the attention and accuracy of individual store associates, not a controlled warehouse environment.
When a shipment arrived, associates were required to log SKU-level details — item identifiers and received quantities — into the company's proprietary store application. If a variance existed between the purchase order and what was actually received, associates also had to manually enter the packing slip quantity and upload a photo of the physical packing slip as supporting documentation. That data then fed an automated debit/credit reconciliation process downstream.
On paper, the process was sound. In practice, it was breaking down in several compounding ways:
Incorrect packing slip quantities were being entered into the system, often because associates were transcribing under time pressure in a busy receiving area. Incorrect or entirely unrelated packing slips were being uploaded as supporting documentation. And because packing slips are physical documents, by the time Finance caught the errors — sometimes days later — the original slip had often been discarded or lost.
The Finance team had implemented a manual post-process audit, reviewing packing slip images in the company's inventory management platform for all variances above a $25 threshold. But this approach came with serious limitations: it was slow, resource-intensive, and almost entirely reactive. Finance couldn't intervene before the data was submitted. Store teams couldn't be reached after the fact. And the audit trail was incomplete.
The operational and financial risks compounded quickly. The company faced the ongoing risk of overpaying vendors due to unvalidated variance claims, delayed supplier payments caused by disputed documentation, and a deteriorating relationship between Finance and store operations — two teams that needed each other to work.
Ashling began with a structured discovery diagnostic to identify the highest-impact automation use cases across the business. The packing slip validation process stood out immediately: high volume, high error rate, significant financial exposure, and a clear technical path to improvement using AI.
The solution Ashling designed centred on a single key insight: the only way to catch a bad packing slip entry is to validate it before it ever hits the downstream system. That meant integrating AI directly into the store associate's existing workflow.
Here's how the PO Variance AI Validation solution’s architecture works:
When an associate completes the receiving process and uploads a packing slip photo through the mobile store application, that data is immediately transmitted to a cloud-based automation flow. An Azure OpenAI model then performs two critical checks: first, it verifies that the image is actually a valid packing slip for the relevant purchase order (filtering out photos of boxes, floors, or unrelated documents); second, it extracts the quantities from the image and compares them against what the associate entered manually.
If the data matches, the variance is automatically cleared in the company's ERP system — no human review required. If the AI cannot validate the image or detects a mismatch, a flag is raised for the Accounts Payable team to investigate. Every result, validated or flagged, is stored in a reporting database for auditing and continuous improvement.
Power Automate served as the orchestration layer, offering a serverless, on-demand execution model that eliminated the need for dedicated infrastructure. Microsoft Azure OpenAI handled document intelligence. And crucially, the solution was built to surface directly within the existing mobile application that store associates already used, boosting adoption of the new flow.
Eighty-five percent of the packing slip variances that Finance previously reviewed manually are now validated and cleared automatically by the AI system. The remaining 15% are flagged and routed to the Accounts Payable team for a targeted review before proceeding. The team that once spent hundreds of hours each week reviewing images, chasing store associates, and reconciling disputed claims has been freed to focus on the exceptions that actually require human judgment.
The project delivered $200,000 in net recurring annual benefit based on time savings alone. That figure doesn't account for the financial risk mitigation from accurately flagging overpayments that previously went undetected. The full economic impact, including avoided vendor overpayments, is meaningfully higher.
Beyond the numbers, the project achieved something harder to quantify: it rebuilt trust between store operations and the Finance team. For the first time, both sides had access to a clear, timestamped audit trail that documented exactly what was submitted, when, and whether it was validated. Disputes could now be resolved with objective data.
The project also proved something strategically important for the company's broader automation journey: that AI could successfully be applied not just as a standalone capability, but as an embedded component of a multi-modal end-to-end automation strategy. That proof point gave leadership the confidence to expand their automation roadmap, with AI-powered IDP now a viable and validated path for tackling higher business process complexity.
What Happens When You Don't Fix the Foundation?
Every month the manual process ran, the company was paying Finance staff to catch errors that could have been prevented at the source — and still missing a portion of them.
Had the retailer not acted, manual review volume would continue to grow as store count scaled. The Finance team would face pressure to add headcount to keep pace, headcount dedicated entirely to reviewing preventable errors. Vendor relationships would continue to strain under disputed payment timelines. And overpayments that slipped through the manual review process would go unrecovered.
Automation didn't just solve a workflow problem. It created the conditions for a different kind of working relationship between two critical functions and a foundation of trust in the data that makes every downstream decision more reliable.